Personalized Recommendation Systems
We retain users and achieve platform business goals through smart recommendations.
The problem we solve
In a highly competitive digital environment, personalization is not just an advantage, it is a necessity. Our recommendation system, developed for a major streaming platform (similar to Kinopoisk, Okko, or IVI), helps retain users by suggesting films and series that match their preferences perfectly. But we go further: the system takes into account not only user interests, but also business goals such as promoting new content or increasing time spent on the platform.
How it works
We built a powerful and flexible recommendation system using modern technologies:
Recommendation Core
At the moment, we use an ALS (Alternating Least Squares) based approach. This choice ensures simple integration and easily interpretable results, which is critical for recommendation quality control.
Automation and Training
Apache Airflow automates the entire pipeline, from data preparation to model training and recommendation generation. This allows the system to refresh recommendations in real time based on the latest user behavior data.
Data and Model Storage
ClickHouse is used as a high-performance database for processing massive volumes of user interactions (~200 GB of view data alone) and provides instant recommendation delivery even under peak load.
API and Filtering
Recommendation requests are handled through a Go API, which guarantees minimal latency. When a request arrives, the system instantly retrieves a list from ClickHouse and applies the necessary filters: excluding already viewed content or movies/series unsuitable for child accounts.
Benefits
Personalized recommendations that keep them coming back again and again.
Higher engagement, stronger retention, and progress toward commercial goals.